Capacity and loyalty models

PhilanthroPy's models are estimators built for fundraising. They follow the scikit-learn estimator interface and add behavior specific to the domain.

Capacity & Potential

Many donors give $1,000 but have the capacity to give $1,000,000. The gap between the two is where major-gift work happens. ShareOfWalletScorer computes a normalized score from 0 to 1 that compares estimated capacity to modeled or known wealth. It tells a gift officer how much room there is to grow a donor's giving.

Gift officers often need to justify the score to leadership. It is defined as:

SoW = predicted_capacity / estimated_total_philanthropic_capacity

This ratio is rooted in industry standards and aligns with benchmarks established by organizations like Blackbaud and EAB.

Capacity Tiers

Categories are easier to act on than raw numbers, so ShareOfWalletScorer.transform emits a second column, capacity_tier, alongside capacity_utilisation_ratio. It is a numeric encoding; get_tier_labels(X) returns the strings.

`capacity_utilisation_ratio`	`capacity_tier`	Label	Recommended action
≥ 0.75	2	Principal	Schedule a personal visit with the campaign chair.
0.40 – 0.75	1	Major	Assign a major gift officer.
< 0.40	0	Leadership	Include in leadership annual giving.

Retention & Loyalty

Lapse Prediction

Find donors at risk of "walking out the back door." Churn models focus on the 12-to-24 month window post-gift, statistically the high-risk "cliff" for donor attrition.

Moves Management

Development officers manage large portfolios of prospects. Moves Management transitions a donor from Identify to Qualify, Cultivate, Solicit, and Steward.

MovesManagementClassifier is not a binary "will they give?" model. It is a multi-class predictive tool that infers a donor's optimal lifecycle stage from recent engagement patterns, which optimizes the workflows of entire fundraising departments.